Supervised learning models can handle time-varying data by using a variety of techniques. One common approach is to use a Sliding Window, where the model is trained on data from a fixed period and then applied to data from a different period. Another approach is to use a Recurrent Neural Network, which can learn temporal dependencies in data. If the input data changes over time, the model can learn to adapt and update its predictions accordingly.

It sounds like you are looking for a model that can learn to classify the fruit based on its appearance, and then can also learn to update its classification based on the specific period. This type of model does not currently exist, but it could be developed with enough data and training.You will need to design your own AI algorithm to accomplish this task.